Reproduce “Known Good, Known Bad: A Ground-Truth Calibration Protocol for AI Judgment in Science” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Olga Lavinda
The authors test a laboratory-evaluation protocol against a verified adipic-acid record and altered versions, checking calculations over turns. In one run per condition, the model detected 3/5 intended deviations initially and all five after a follow-up prompt. It recognized a liters-to-milliliters error but retained an incorrect limiting-reagent conclusion. This pilot covers one case family, not general performance or a causal effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.